Anthropic has been touting the power of its Claude Mythos AI model to find and help fix security vulnerabilities, and it has finally released a public version, Fable 5.However, some cybersecurity experts are complaining that the guardrails meant to prevent abuse are stifling its real-world usefulness.Fable is billed as a "Mythos-class" model that can conduct reasoning, coding, and cybersecurity tasks better than both the Mythos preview release and conventional AI models, including OpenAI's GPT 5.5 and Google's Gemini 3.1 Pro.
This ideally helps security teams pinpoint flaws, study issues, and develop fixes with relative speed.Claude Price $20 See at Claude Expand Collapse The differences between Fable 5 and its Mythos 5 equivalent come down to the "safeguards," according to Anthropic.If the Fable model believes there's a risk of abuse, such as writing malware, it reverts to Claude Opus 4.8.
Mythos 5 doesn't have as many protections, and access is limited to members of the Project Glasswing initiative meant to secure vital software with the help of the U.S.government and certain companies, including Apple, Google, Microsoft, and hundreds of others.The new Fable and Mythos versions are available at respective prices of $10 and $50 per million input tokens, or less than half the price of the Mythos preview.
Pros who want fewer limits on any of Anthropic's models have to get approval through a Cyber Verification Program.Why is Fable 5 so bad for some cybersecurity experts? They feel it's simply too safe Anthropic is clear that it set Fable's guardrails "conservatively" for the sakes of both a quick launch and safety, and that there will be moments where "harmless requests" limit functionality.It expects to improve these protections and limit mistakes going forward.
As explains, however, cybersecurity veterans still aren't thrilled.Experts at IBM and elsewhere have pointed out that asking to write or review secure code downgrades to Opus, and that the limits might kick in simply by using the 'wrong' keyword.For some, it could be difficult to use Fable the way it was intended.
Related 5 AI Technologies Hackers Can Use in Terrible New Ways Some people think The Matrix is an instruction manual.Posts By Sydney Butler The initial caution might be warranted.AI can discover and use vulnerabilities more easily and quickly than before, and that could lead to zero-day exploits that developers haven't had time to fix or even analyze.
Subscribe to our newsletter for AI security insights Want expert context on AI security? Subscribe to our newsletter for in-depth coverage of model guardrails, vulnerability risks, and how AI tools shape cybersecurity.Coverage breaks down technical trade-offs and clarifies the stakes for defenders and developers.Get Updates By subscribing, you agree to receive newsletter and marketing emails, and accept our Terms of Use and Privacy Policy.
You can unsubscribe anytime.Microsoft, for instance, marked a record "Patch Tuesday" in June where it fixed almost 200 security bugs as its team and the community used AI to catch issues they'd otherwise miss.If attackers could use tools like Fable or Mythos without limits, there's a risk they could routinely compromise systems before patches are ready, and with relatively little effort.
Read More
